Integrated water-quality sensors
The boat can carry in-water sensors alongside its camera, so assessment doesn't stop at what's visible on the surface. Sensor readings and visual findings land in the same geo-tagged report.
AUTONOMOUS SURVEY BOAT · AMSTERDAM
An autonomous robot boat that watches over Amsterdam's canals with AI.
Compact and 3D-printed, it patrols the canals on its own — filming the water, measuring it with onboard water-quality sensors, and turning every route into a clear report.
WHY A ROBOT
100+ km of canals. Occasional inspections can't keep up.
Amsterdam's canals are well kept, yet local problems still surface: floating waste, algal blooms, oil sheens. Across a network this large, catching them early is slow and expensive.
Autonomous boats remove the bottleneck. Several can run at once and cover the whole network — measuring far more densely, in space and in time.
TECHNOLOGY
A floating computer. Built to find its own way.
Canals are busy: moving boats, floating objects, sudden obstacles. The boat fuses its sensors through ROS2 and steers itself along a given route with AI.
Stereo camera
RGB plus depth. Sees obstacles in 3D and films the water for analysis.
GPS
Every frame is stamped with time and position, so every finding has an address.
Accelerometer & magnetometer
Heading and motion, for stable navigation between GPS fixes.
ROS2
One robotics stack ties sensors, control and logging together.
Onboard navigation model
A custom, lightweight model steers past traffic and obstacles in real time.
3D-printed hull
About 1.5 m long. Small enough for narrow canals, cheap enough to build a fleet.
- 01
Navigate
Follows a route on its own, avoiding obstacles.
- 02
Capture
Video, frames and sensor data, all time- and geo-stamped.
- 03
Analyse
A local vision model reads the footage, frame by frame.
- 04
Report
One concise page: issues, locations, severity.
RIDE ALONG · SHORT CUT
A few minutes from a ride that lasts hours and covers kilometers.
This is a short excerpt of one multi-hour survey through Amsterdam-Noord — onboard-camera frames, each with its GPS fix, played back with the route. Press play or drag the slider.
frame 1 / 602
17:25:57
52.40728, 4.92066
ROUTE REPORTS
Hours of footage in. One page out.
Even one route produces hours of video and sensor readings. Reviewing that by hand is a bottleneck of its own — so the boat's footage is analysed automatically and condensed into a short report.
- Detected issues, by type
- Exact locations on the map
- Severity levels
- Plain-language observations
GROUNDED IN SCIENCE
Judgement you can explain.
Severity isn't a vibe. Every assessment is anchored in established environmental frameworks.
- WHO 2021 Recreational Water Quality
- Bonn Agreement Oil Appearance Code
- OSPAR macro-litter monitoring
- Wet milieubeheer Art. 10.2
- EU Bathing Water Directive 2006/7/EC
Objective criteria
Raw vision models give subjective scores. Grounding them in real standards makes reports consistent and comparable.
Context, not just objects
A bicycle on the quay is Amsterdam. A bicycle in the canal is pollution. The model reads the scene — and isn't limited to a fixed list of classes.
Plain language
“A bulky human-made object, partially submerged” is now simply “a shopping cart”. Reports are written for people.